Testing Your Agent

1# part3/testing_agents.py
2import json
3from openai import OpenAI
4client = OpenAI()
5
6# --- The agent under test: a calculator with one tool ---
7def add(a, b):
8    return a + b
9
10tools = [{"type": "function", "function": {
11    "name": "add",
12    "description": "Add two numbers together",
13    "parameters": {
14        "type": "object", "required": ["a", "b"],
15        "properties": {"a": {"type": "number"},
16                       "b": {"type": "number"}}},
17}}]
18
19def ask_agent(question):  # returns (tool_used, answer)
20    messages = [{"role": "user", "content": question}]
21    response = client.chat.completions.create(
22        model="gpt-4o-mini", messages=messages,
23        tools=tools, temperature=0)
24    message = response.choices[0].message
25    if not message.tool_calls:
26        return None, message.content  # no tool used
27    tool_call = message.tool_calls[0]  # asked for add
28    arguments = json.loads(tool_call.function.arguments)
29    result = add(arguments["a"], arguments["b"])
30    messages.append(message)
31    messages.append({"role": "tool",
32                     "tool_call_id": tool_call.id,
33                     "content": str(result)})
34    final = client.chat.completions.create(
35        model="gpt-4o-mini", messages=messages,
36        temperature=0)
37    answer = final.choices[0].message.content
38    return tool_call.function.name, answer
39
40# --- Tests: a question + what a good answer must have ---
41tests = [
42    {"question": "What is 45 + 13?",
43     "expect_tool": "add", "must_contain": "58"},
44    {"question": "What is 250 + 175?",
45     "expect_tool": "add", "must_contain": "425"},
46    {"question": "What is the capital of France?",
47     "expect_tool": None, "must_contain": "Paris"},
48]
49
50# --- Run every test, check it, and keep score ---
51passed = 0
52for test in tests:
53    tool_used, answer = ask_agent(test["question"])
54    print(f"\nQ: {test['question']}\nA: {answer}")
55    tool_ok = tool_used == test["expect_tool"]  # check 1
56    answer_ok = test["must_contain"] in answer  # check 2
57    if tool_ok and answer_ok:
58        print("✅ PASS")
59        passed += 1
60    if not tool_ok:
61        print("❌ FAIL: expected tool "
62              f"{test['expect_tool']}, "
63              f"but it used {tool_used}")
64    if not answer_ok:
65        print("❌ FAIL: the answer should contain "
66              f"'{test['must_contain']}'")
67
68print(f"\nScore: {passed}/{len(tests)} passed")